Frequently Asked Questions: The Intrepid Platform and Generative AI.
AI technology is evolving rapidly, but at Intrepid, our commitment to the responsible use of AI and serving learners and industry partners remains unwavering.
Whether we're enhancing learning experiences for participants or optimizing solutions for L&D professionals, our guiding principles shape our approach to integrating AI and learning. If you have a question that's not listed here, please reach out to us.
Whether we're enhancing learning experiences for participants or optimizing solutions for L&D professionals, our guiding principles shape our approach to integrating AI and learning. If you have a question that's not listed here, please reach out to us.
What technology does Intrepid use for the GenAI features on the Intrepid Platform?
The Intrepid Platform uses OpenAI’s APIs to access the GPT 3.5 and GPT 4.0 models.
How is personally identifiable information (PII) protected?
Intrepid will never pass any PII fields that are in the Platform to OpenAI: e.g., first/last name, username, or email address. However, Intrepid does not control the information (i.e., text) that an administrator or learner may choose to enter into the Platform’s GenAI features. Clients should establish and communicate policies about the data that users are able to enter into the Intrepid Platform GenAI text fields.
What other technology partners does Intrepid use to implement the Platform’s GenAI features?
Intrepid partners with Tiny Technologies to provide the TinyMCE html editor on the Platform. Through TinyMCE, Intrepid offers the Intrepid AI Editor.
Intrepid partners with Bongo Learn to provide the Interactive Video Assignment feature on the Platform. The Interactive Video Assignment feature has GenAI functionality, specifically the AI Coach and Smart Scoring features.
Does TinyMCE send data directly to OpenAI?
No, TinyMCE does not send data directly to OpenAI. The text that an administrator chooses to use in the Intrepid AI Editor (either through highlighting existing text or adding new text in the prompt) is sent to OpenAI via the Intrepid Platform’s secure proxy. TinyMCE does not see the data.
Does Bongo Learn send data directly to OpenAI?
Bongo sends video transcript data to OpenAI. It does not send the assignment video. Bongo has an FAQ with more details of their GenAI functionality, which you can find here.
How is data used by OpenAI?
OpenAI does not use any data from the Intrepid Platform (including your Client Data) to train or improve models.
In order to retrieve coaching tips and tonal analysis, Bongo sends the video transcript and Auto Analysis metrics. The actual video and any information about the user are not sent. This content is not used to train any language models to avoid exposure to proprietary or private information.
Data is encrypted in transit and at rest. OpenAI is audited for SOC 2 Type II compliance.
Data is retained for 30 days by OpenAI, then it is removed. For more information see https://openai.com/enterprise-privacy.
Are Intrepid’s GenAI features on by default?
No. All GenAI features on the Intrepid Platform are disabled by default. If you want to use Generative AI features, please contact your Intrepid Account Manager.
I’m an admin, how do I request Intrepid’s AI features be turned on for my classes?
If you want to use Generative AI features, please contact your Intrepid Account Manager.
Is the data collected retained? If so, for how long and why?
Data is retained for 30 days by OpenAI, then it is removed. For more information see https://openai.com/enterprise-privacy. Customer and/or user data entered into the Bongo system that is utilized in AI offerings is retained by the model for 30 days for audit purposes after which it is automatically purged from the system.
How is data poisoning, hallucinations, and bias managed?
The Intrepid Platform uses OpenAI’s APIs to access the GPT 3.5 and GPT 4.0 models. No AI-generated content is published directly to learners without admin oversight. Intrepid encourages clients to use generative AI features to augment the content creation processes, not as a complete replacement for human/admin involvement.
The Bongo team is aware that AI models can be subject to bias based on the data set on which they have been trained. While there is no magic bullet, Bongo utilizes safety measures. First, closely controlling how the OpenAI model is prompted ensures more consistent results as compared to an open-ended experience. In the first release and until further testing confirms reliability, Bongo uses a separate service to transcribe the speech in the video and does not send the actual audio track to provide an objective assessment.
The Bongo team is aware that AI models can be subject to bias based on the data set on which they have been trained. While there is no magic bullet, Bongo utilizes safety measures. First, closely controlling how the OpenAI model is prompted ensures more consistent results as compared to an open-ended experience. In the first release and until further testing confirms reliability, Bongo uses a separate service to transcribe the speech in the video and does not send the actual audio track to provide an objective assessment.
What is the protection against model inversion?
The Intrepid Platform uses OpenAI’s APIs to access the GPT 3.5 and GPT 4.0 models. OpenAI does not use any data from the Intrepid Platform (including your Client Data) to train or improve models. Bongo utilizes third-party large-language models.
Model inversion protection is the responsibility of the model providers. That being said, Bongo’s strategy is to interact with the models through API services that do not train the models. This avoids risk that Client Data and/or PII entered into Bongo could be compromised through an inversion attack.
Model inversion protection is the responsibility of the model providers. That being said, Bongo’s strategy is to interact with the models through API services that do not train the models. This avoids risk that Client Data and/or PII entered into Bongo could be compromised through an inversion attack.
What personal information is collected and sent to OpenAI?
Intrepid will never pass any PII fields that are in the Platform to OpenAI: e.g., first/last name, username, or email address. However, Intrepid does not control the information (i.e., text) that an administrator or learner may choose to enter into the Platform’s GenAI features. Clients should establish and communicate policies about the data that users are able to enter into the Intrepid Platform GenAI text fields.
For the Interactive Video Assignment feature using Bongo technology, no PII is sent to OpenAI's GPT model. Only data related to the video transcript generated by AWS Transcribe is sent to OpenAI. The transcript data is encrypted both in transit and at rest. Bongo sends the data to a version of the model that is appropriately located relative to a user’s region. Finally, by using OpenAI models through an API, the transcript is not used for training the models, thereby avoiding exposure to proprietary or private information.
For the Interactive Video Assignment feature using Bongo technology, no PII is sent to OpenAI's GPT model. Only data related to the video transcript generated by AWS Transcribe is sent to OpenAI. The transcript data is encrypted both in transit and at rest. Bongo sends the data to a version of the model that is appropriately located relative to a user’s region. Finally, by using OpenAI models through an API, the transcript is not used for training the models, thereby avoiding exposure to proprietary or private information.